How Google’s West Virginia Solar Project Addresses PJM’s Power Gap

A Google data center is about to be powered by a coal mine in West Virginia. That sentence contains three things that once seemed structurally incompatible: a hyperscaler’s clean energy pledge, a piece of Appalachian mining infrastructure, and a grid that can supply neither on its own.

None of this is symbolic. The PJM Interconnection, the grid that serves this region, is short roughly 6.8 GW of firm capacity, its capacity auction has cleared at the regulatory price ceiling three years running, and hyperscalers like Google need power that does not yet exist on the grid.

The Kanawha County solar and storage project is a structural answer to structural scarcity. It sits at the intersection of corporate energy strategy, grid reliability policy, and post-industrial land use.

Understanding how those three layers connect is the fastest way to understand how American electricity actually works in an AI era, and why the West Virginia solar storage deal is being watched far beyond West Virginia.

What Google and MN8 are actually building in Kanawha County

The facility rising on a retired coal mine site in Kanawha County, West Virginia, is not a solar farm. It is a three-part power system, and each part is doing a different job.

The generation layer is 86 MW of solar photovoltaic capacity, with commercial operation projected for 2028. On its own, that is a conventional utility-scale array. What makes this project unusual is what sits alongside it.

Data center load curves create a procurement challenge that conventional grid connections were not engineered to serve: a near-flat, round-the-clock demand profile that spikes during peak compute cycles and cannot absorb the intermittency that simple renewable energy contracts deliver.

The developer, MN8 Energy, has paired the solar with two chemically distinct storage systems. That choice is the whole point, and it is worth slowing down on.

Why two storage chemistries instead of one

Lithium-ion storage is the short-cycle workhorse. It delivers high power, dispatches fast, and handles the sudden ups and downs of a data center load curve, but it discharges over a relatively short window. The project’s lithium-ion system is rated at 70 MW / 280 MWh, with commercial operation projected for 2029-2030.

The second system solves a different problem. Once the solar day ends, something has to keep power flowing through the overnight hours, and lithium-ion alone cannot economically hold that long a discharge. That is the job of the zinc-based long-duration battery supplied by Eos Energy Enterprises, rated at 10 MW / 100 MWh and also projected for 2029-2030.

Zinc-based storage occupies a distinct commercial position from lithium-ion: it sacrifices peak power delivery in exchange for longer discharge windows and a chemistry that avoids the thermal runaway risks that complicate lithium deployments at grid scale.

Zinc-based storage delivers lower power but far longer duration, making it suited to the multi-hour, overnight backstop role. Pairing the two chemistries lets the facility track a data center’s demand through the full 24-hour cycle rather than just the sunny part of it.

The two-chemistry stack tells you this was engineered for reliability, not clean-energy optics. Google’s arrangement covers energy output, capacity credits, and clean energy certificates, so this is a bundled procurement, not a simple electricity purchase.

That distinction matters, because the capacity credit is the piece that speaks directly to what PJM’s market is currently failing to supply.

The PJM capacity crisis that makes this project matter

To understand why Google is buying capacity credits and not just electrons, you need to understand what PJM’s capacity auction has been telling the market for three years.

PJM runs a Base Residual Auction (BRA) to secure enough firm generating capacity to keep the lights on years ahead of time. Generators bid, the auction clears at a price, and that price is meant to signal whether supply is comfortable or tight. When an auction clears at its regulatory ceiling, that is a distress signal, not a normal outcome.

PJM has now cleared at or near that ceiling three delivery years in a row. The 2026-2027 BRA cleared at $329.17/MW-day. The 2027-2028 auction cleared at $333.44/MW-day and left the system 6,517 MW short of its reliability requirement. The 2028-2029 auction cleared at the FERC-approved price cap of $325/MW-day across the entire PJM footprint.

Three consecutive years at the ceiling is not a pricing quirk you can wave away. It is the market saying, as plainly as a market can, that the region cannot build new firm capacity fast enough to keep up with load growth.

The grid capacity crisis unfolding across PJM is not an isolated market quirk; it reflects a structural mismatch between the pace of load growth driven by data centers and AI, and the pace at which new firm capacity can clear permitting, interconnection, and construction timelines.

The 2028-2029 shortfall reached 6,831 MW, leaving PJM with a realised reserve margin of roughly 14.7% against a target of around 20%. That is the gap the entire backstop mechanism now exists to close.

Because the standard auction could not close that gap, PJM has scheduled a one-time supplemental Reliability Backstop Procurement (RBP). It is designed as a two-stage, pay-as-bid, capacity-only auction targeting shovel-ready new-build resources, with a proposed price cap of $555/MW-day.

The RBP bid window runs from 30 September to 21 October 2026, with results expected between 22 October and 2 December 2026. Contract terms range from 2 to 15 years, and selected resources must reach commercial operation by 1 June 2031.

That backstop design is exactly why a project like Kanawha County is being structured to earn capacity credits rather than energy revenue alone. Seen against these numbers, the Google-MN8 deal stops looking like a public-relations gesture and starts looking like a calculated response to a reliability problem regulators have openly admitted the standard market cannot fix.

What it actually takes to build solar on a retired coal mine

The instinct is to treat the coal site as pure upside: free land, an industrial history, existing infrastructure, and a clean-energy story that writes itself. The engineering reality is more complicated, and it reshapes how you should read the timeline.

A retired mine is not a blank field. It is physically dynamic ground with a set of specific hazards that have to be neutralised before a single solar tracker goes in.

The core technical challenges specific to mine-land development include:

• Subsidence: Mined-out ground can shift horizontally and vertically, threatening solar tracker foundations and electrical equipment, which forces rigorous geotechnical and residual-subsidence assessments.
• Acid mine drainage: Toxic drainage from abandoned workings must be remediated as a reclamation prerequisite before construction can begin.
• Soil instability: Erosion and unstable ground require extensive site preparation.
• Transmission gaps: Remote Appalachian mine sites often lack adequate transmission, making grid upgrades one of the costliest hurdles to new entry.

Each of those items adds engineering time, and that is the honest explanation for the 2028-2030 build sequence. The timeline is not slow execution. It reflects the real regulatory and engineering steps required to develop safely on brownfield mine land, and it tells you precisely when this capacity will actually show up in PJM’s reliability math: solar in 2028, storage across 2029-2030.

Where former coal sites give developers an edge

The site is not only a liability. Former power plant locations, as distinct from remote strip mines, frequently carry existing high-voltage interconnection, and interconnection is one of the single most expensive bottlenecks for new solar entry in PJM. Inheriting a grid connection can save years and substantial cost.

The political context helps too. West Virginia legislators and federal programmes, including Inflation Reduction Act energy community bonus credits aimed at former fossil-fuel areas, create an incentive alignment that can streamline permitting relative to a greenfield build.

The IRS energy community bonus credit guidance clarifies that mine-scarred land may qualify as a brownfield site under 42 U.S.C. 9601(39)(A), making Kanawha County-type projects eligible for additional Inflation Reduction Act incentives that can meaningfully offset the higher upfront costs of mine-site development.

Industry precedent suggests a 3 to 4 year path from policy framework to commercial operation for coal-site repurposing, which lines up with MN8’s schedule. Comparable transitions exist elsewhere: Duke Energy is reported to have brought a 50 MW / 200 MWh battery system online at the retired Allen coal plant in January 2026, though that figure carries a lower verification confidence and should be treated as directional. The pattern across successful projects is consistent: they clear the environmental work first, then build.

How hyperscalers are rewriting the rules of regional power procurement

Zoom out from Kanawha County and a larger shift comes into focus. Hyperscalers are no longer taking what the grid offers. They are commissioning the grid’s next chapter.

For years the standard corporate clean-energy playbook was the virtual power purchase agreement: a financial contract tied to off-site intermittent renewables that offset consumption on paper without delivering firm power to a specific load. That model is giving way to something more physical.

• The old model: Virtual PPAs, off-site intermittent renewables, energy-only contracts that offset consumption without guaranteeing round-the-clock delivery.
• The emerging model: Physical delivery contracts, co-located or bundled assets, and firm capacity included, engineered to match a data center’s 24/7 load.

Google’s Kanawha County deal is a live example of the second model, and it fits inside a far larger regional bet. Google has flagged an estimated $25 billion over two years in data center and AI infrastructure across PJM states, plus $3 billion to modernise two Pennsylvania hydropower plants.

The scale of what hyperscalers are now commissioning is visible in Google’s separate Minnesota project with Xcel Energy and Form Energy: a 300 MW / 30 GWh iron-air battery capable of up to 100 continuous hours of discharge. These are grid-shaping assets, not corporate offsets.

Google has also pushed a “Clean Transition Tariff” concept, an attempt to move the hyperscaler-utility relationship beyond one-off PPAs and into formal grid planning. Industry estimates suggest the four largest hyperscalers accounted for roughly 49% of global clean-energy PPA volumes in 2025, a figure worth treating as directional given its lower verification confidence, but one that captures how much of the market these buyers now drive.

There is a policy edge to all of this, and it lands directly on procurement decisions.

Under the proposed RBP cost allocation, backstop procurement costs are directed to load-serving entities that serve new large data centers that have not self-procured capacity or agreed to curtailment. In plain terms, hyperscaler loads that fail to arrange their own firm capacity will help pay for the grid’s backstop.

That rule reframes the Kanawha County deal. Among everything else it is, it is also a hedge. By procuring capacity credits directly, Google positions itself on the covered side of that cost-allocation line rather than the uncovered side.

The broader takeaway is that hyperscaler procurement is now actively deciding which resources get built, where they land, and who pays for the reliability infrastructure behind them.

What this project signals, and what still has to go right

The Kanawha County facility is real and contracted, but it does not generate a megawatt-hour until at least 2028, with storage following in 2029-2030. PJM’s 6,831 MW shortfall is a problem for now. The distance between those two facts is where the reliability story actually lives.

The project matters on three fronts at once. It is a grid reliability play built to earn capacity credits, a coal-to-clean transition demonstration on genuinely difficult ground, and a hyperscaler procurement template that other developers and utilities will study closely. It sits inside Google’s roughly $25 billion PJM regional commitment, which gives it staying power.

Firm capacity under grid stress performs differently from nameplate ratings suggest: the Winter Storm Fern episode demonstrated that resources able to sustain output through sustained demand spikes command an entirely different reliability value than those that ramp down when the grid needs them most.

Whether it delivers its intended grid value comes down to a short list of variables worth tracking:

1. The autumn 2026 RBP outcome. Bids close 21 October 2026 and results arrive by 2 December 2026. That result will tell you whether the market has enough shovel-ready supply to close even part of the 6.8 GW gap, and it will land before Kanawha County produces anything.
2. Execution against the 2028-2030 timeline. Mine-site engineering, subsidence work, and acid mine drainage remediation all have to go to plan for the capacity to show up when the reliability math needs it.
3. PJM’s structural demand trajectory. If data center load keeps outrunning new supply, one project, however well designed, will not move the aggregate.

The larger question this deal forces open is whether private hyperscaler procurement, however sophisticated, can substitute for the systemic grid investment PJM’s capacity market has so far failed to incentivise on its own. Kanawha County is one strong answer. It is not the whole answer.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions.

Forward-looking statements regarding project timelines, auction outcomes, and capacity delivery are speculative and subject to change based on market developments, regulatory decisions, and project execution.